Forecasting of Nuclear Power Plant Burst Factors using Radial Basis Neural Network throughout Loss of Coolant Accident

This research aimed to improve the prediction of nuclear fuel claddings' characteristics in burst trials under simulated loss-of-coolant accident (LOCA) events. The established burst criterion was found to have limitations, including assumptions of symmetrical deformation, mixed-phase cladding properties, and disregard of azimuthal temperature changes. To address these limitations, a radial basis neural network (RBNN) model was developed using a backpropagation technique and a Gaussian radial basis activation function in the single hidden layer. The model demonstrated superior prediction capabilities, with average variances of $0.358 %, 0.082 %$, and $0.067 %$ for temperature, stress, and strain, respectively. The model was also evaluated with noisy input data. The suggested RBNN model was found to outperform both traditional systems and other developed artificial neural network (ANN) models in terms of effectiveness and efficiency.